daily_stock_analysis hands-on: a failed fetch became '0' market volume

The tool itself is real: the offline unit tests passed 8,476 cases in about 12 minutes, and data fetching routed correctly per market, pulling actual recent closes such as 600519's 1453.00/1444.99/1439.00 yuan over three sessions. The problems surfaced only when the free, local path was actually exercised.

The market review is where the design gap shows. Six data sources failed in sequence -- a JSON parse error, two 30-second Akshare timeouts, and four providers with no matching data -- yet the prompt handed the model only the number 0, with no context that the fetch had failed. Individual stock dashboards, by contrast, carry an explicit data-limitation field listing missing volume, chip-structure and quote data, so the omission appears specific to the market-review path.

Output stability is the other finding. On the local 7B model, schema validation errors recurred and fallback filling kept the pipeline from stopping, which is why the run was counted as a success despite placeholder content. The later stall of the second market review, while Ollama stayed responsive to a direct curl, is recorded as reproducible but not diagnosed in the source code.